Multi-extended Target Tracking Based On Gaussian Process And PHD Filtering Algorithm

Nowadays,the accuracy of the sensor has been greatly improved.The target no longer produces only one measurement point under the sensor,but multiple,which makes the traditional point target tracking technology is no longer applicable,and the research on extended target tracking has been widely concerned,but in practical application,multi-extended target tracking is more common.Therefore,it is more meaningful to study multi-extended target tracking.Multiple extended targets make the data association between targets and measurements more complex and increase the difficulty of research.With the introduction of random finite set theory into the filtering framework of multi-target tracking,the complex data association between targets and measurements is effectively avoided.On the basis of this framework,this paper studies the problem of multi-extended target tracking with different shapes by fully considering the extended state of the target.Aiming at the problem of shape estimation of star-convex irregular targets,a shape modeling method of Gaussian process regression is adopted in this paper.In this method,the target shape is described by the mean and covariance function of the radial function,and the target shape is updated by means of Kalman recursion.By combining this modeling method with multi-extended target probability hypothesis density(PHD)filter,the motion state and shape state of extended targets can be estimated effectively at the same time.The estimation accuracy of motion state and shape state of multi-extended targets is evaluated by OSPA distance and quasi-Jaccard distance.Finally,through the simulation comparison,it is verified that the algorithm has a better outline estimation effect than the star convex random hypersurface modeling,and can realize the effective tracking of the target trajectory.In order to solve the problem of multi-extended target tracking in nonlinear system,a sequential Monte Carlo probability hypothesis density(SMC-PHD)filtering algorithm based on Gaussian process is adopted.Firstly,an improved firefly clustering algorithm is used to solve the problem of low accuracy of target motion state estimation in SMC-PHD filtering.The clustering algorithm does not depend on the selection of the initial center of clustering,and guides other fireflies to combine and update the optimal fireflies through the attraction and movement mechanism of the optimal fireflies,so as to obtain the multi-extended target state.Secondly,combined with the Gaussian process regression modeling method model,the motion and shape states of multi-extended targets are jointly estimated.Finally,the effectiveness of the algorithm for object motion state and extended state estimation is verified by multi-extended object simulation experiments.